{"id":166695,"date":"2026-03-30T11:13:41","date_gmt":"2026-03-30T09:13:49","guid":{"rendered":"https:\/\/prohoster.info\/blog\/novosti-interneta\/anthropic-predupredila-chto-eyo-budushhaya-model-claude-mythos-stanet-hakerskoj-vundervaflej"},"modified":"2026-03-31T11:20:25","modified_gmt":"2026-03-31T09:20:25","slug":"anthropic-predupredila-chto-eyo-budushhaya-model-claude-mythos-stanet-hakerskoj-vundervaflej","status":"publish","type":"post","link":"https:\/\/prohoster.info\/en\/blog\/news\/anthropic-predupredila-chto-eyo-budushhaya-model-claude-mythos-stanet-hakerskoj-vundervaflej","title":{"rendered":"Anthropic has warned that its upcoming model, Claude Mythos, will become a \"hacker's wonder weapon.\"","gt_translate_keys":[{"key":"rendered","format":"text"}]},"content":{"rendered":"

Anthropic, OpenAI, and other technology companies are preparing AI models capable of significantly increasing the threat of large-scale cyberattacks on corporate, government, and municipal systems. The focus is on the yet-to-be-released AI model Claude Mythos from Anthropic. According to Axios, the company is already warning high-ranking U.S. officials that its emergence in 2026 significantly raises the likelihood of such attacks.<\/p>\n

<\/center>\"Anthropic<\/p>\n

Mythos is described as a system that allows AI agents to autonomously infiltrate corporate, government, and municipal systems with high precision and complexity. Axios CEO Jim VandeHei reported in his newsletter for executives that one source familiar with future AI models anticipates a major cyberattack as early as 2026, with businesses being noted as particularly vulnerable targets.<\/p>\n

Earlier, Fortune studied<\/a> leaked materials from Anthropic's blog about Claude Mythos. In it, the AI model is described as a system that \"currently significantly outpaces any other AI model in cyber capabilities.\" It also states that Mythos \"forecasts an impending wave of AI models capable of exploiting vulnerabilities in ways that will far outpace cybersecurity efforts.\" The risk is exacerbated by employees using AI agents without understanding that this may simplify access for cybercriminals to their companies' internal systems.<\/p>\n

At the end of 2025, Anthropic revealed<\/a> documented the first case of a cyberattack largely executed by AI. A state-backed Chinese group used AI agents for the autonomous hacking of approximately 30 targets worldwide. AI independently conducted 80% to 90% of the tactical operations. This occurred even before the significant enhancement of agent systems and the emergence of new hazardous methods of their application.<\/p>\n

New AI models exhibit a greater ability to maintain the autonomous operation of AI agents. These systems allow agents to think, act, reason, and improvise independently without breaks or fatigue. The scale of the attack is now primarily determined by the computational resources of the attackers rather than the size of their group. Now, one person can achieve of<\/a> conduct cyberattacks that previously required entire teams.<\/p>\n

At the same time, the vulnerability of companies is increasing. Employees are increasingly launching Claude, Copilot, and other agent models, often from home, and creating their own AI agents. In many cases, these agents gain access to internal work systems without proper control. This creates another channel for cybercriminals to exploit. In the industry, this unauthorized or uncontrolled use of AI is referred to as 'shadow AI'.<\/p>\n

A survey by Dark Reading revealed that 48% of cybersecurity professionals believe agent AI is the main attack vector in 2026. This threat is rated higher than deepfakes and all other risks included in the survey. Every employee in any company worldwide needs to understand the risks of using AI agents alongside sensitive information. Axios's technical team considers this threat to be the biggest for companies at present. The practical conclusion is clear: a secure isolated environment is necessary for working with AI agents.<\/p>\n

Source:<\/strong><\/p>\n